KR950002705B1 - Speech synthetic system - Google Patents

Abstract

The syntactic parser analyzing the semantic relations of the written text statement to control the intonations for speech synthesis comprises a file input means(1) reading the text statement stored in an auxiliary memory device; a morphology analyzing means(2) finding the structure of the text statement; a dictionary means(3) keeping the information for analyzing the morphology; a syntactic handler(5) processing the text statement in the phrase; a paragraph/phrase handler(4) separating phrase parts from a paragraph; an intonation handler(7) separating the predefined intonation sections; an intonation extractor generating intonation information for speech synthesis.

Description

Parsing System and Method for Rhythm Control in Speech Synthesis System

1 is a flowchart of speech synthesis processing of a conventional speech synthesis system.

2 is a speech synthesis processing flow diagram of a speech synthesis system having a syntax analysis device of the present invention.

3 is a block diagram of a syntax analysis device of the present invention.

4 is a syntax analysis flowchart showing a syntax analysis process of the present invention.

(A) of FIG. 5 is a flow chart for each word preprocessing according to the present invention, and (b) is a diagram showing the type of dictionary according to the present invention.

6A is a diagram showing an example of a structure of an information frame according to the present invention, and (B) is a table showing the eigenvalues and the meanings of the eigenvalues of the information according to the present invention.

7 is a flow chart of the parsing word processing process of the present invention.

8 is a flowchart of a noun sequence processing of the present invention.

9 is a flow chart of the predicate processing of the present invention.

10 is a flow chart of a connection inquiry connection word processing process of the present invention.

11 is a flowchart of a syntax indicator processing process of the present invention.

12 is an example of parsing processing output of the present invention.

(A), (b) and (c) of FIG. 13 are examples of application of intonation to the sentence of the present invention.

14 is a flow chart of the morphological analysis process of the present invention.

15 is a flow chart of the syntactic analysis process of the present invention.

16 is an explanatory diagram of terms and symbols used in each process of the present invention.

* Explanation of symbols for main parts of the drawings

1: File input means 2: Morphological analysis means

3: dictionary means 4: syntactic analysis means

5: syntax rule processing means 6: phrase separation means

7: accent separation means 8: rhyme information extraction means

The present invention relates to a syntax analysis device and a syntax analysis method for synthesizing a speech through a rhythm control by analyzing a sentence to be synthesized in a speech synthesis system for receiving an input sentence and synthesizing it into a speech. The present invention relates to a syntax analysis device and a syntax analysis method for rhyme control of a Korean sentence / voice converter, which improves the naturalness and clarity of the synthesized sound by controlling Korean rhythm and the result of the analysis.

As shown in FIG. 1, the functional configuration of a conventional speech synthesis system includes a sentence preprocessor for processing long sounds or irregular speech fluctuations of input sentences, a rhythm implementation and a phonological phenomenon of the preprocessed sentences. And a voice synthesizer for synthesizing and outputting the rhyme and the phonologically processed sentences into voices. The voice synthesis process is as follows.

The sentence preprocessing unit performs a phonological process of Korean on the input Korean sentence, and converts the Korean sentence into a phonetic sound as a result of the processing.

The sentence processing information required by the sentence processing unit is provided in a long sound dictionary, an irregular phonological change dictionary, an investigation, and a parent dictionary, and the sentence preprocessing unit is based on the information provided in each dictionary. The processing results are supplied to the rhythm realization and phonological processing section.

Rhyme implementation and phonological phenomenon processing unit implements rhyme of Korean by processing phonological phenomena and stress, syllable length, and intonation within a word according to the structure of Hangul phonemes using the result of the sentence preprocessor.

The information necessary for processing the rhyme and phonological phenomena is constructed by providing the pitch trajectory information as a database (PITCH trajectory DBASE) and providing this information, and the energy information of the rhyme implementation is also constructed as the data bears (ENERGY trajectory DBASE). To provide.

Using the information from the database, the sentence processed by the rhyme implementation and phonological development processing unit is supplied to the speech synthesis unit.

The voice synthesis unit outputs the voice corresponding to the input sentence by making the input sentence information into sound according to the voice data provided by the voice database (voice DBASE).

In other words, the voice database stores the voice data in units of phonemes, syllables or half-syllables, and supplies them to the voice synthesizer, and the voice synthesizer combines them according to the input sentence and makes them into sound. It approximates natural sound.

However, the conventional speech synthesis system as described above has a problem in that the synthesized sound is not natural and the naturalness is lowered because the interpretation of the sentence of Korean is not preceded.

In other words, the rhyme of Korean language is divided into two sections: accents, accents, and syllable lengths, which are preconditions between sentences and sentences. Among them, stress and syllable lengths are composed of phonemes. Regular control is possible with values determined according to the combination of neutral and final.

However, the word rest period and the adjustment of the accent could not be adjusted as naturally as the human voice.

Of course, although the minimum unit of accent is applied to each word, the rest period in accent and sentence is calculated by the semantic, syntactic, and segmental characteristics in sentence units, and cannot be applied by word and syllable units. .

Therefore, if the interpretation of the sentence is not preceded, the quality of the synthesized sound is difficult to expect beyond the monotonous mechanical sound, which in turn leads to a problem in that it fails to meet the increasing demand of human machine information transmission such as MULTIMEDIA. Like this, there is a problem of failing to realize natural high quality synthesized sound.

An object of the present invention is to find a grammatical relationship between words for a sentence to be synthesized, and to provide syntax analysis means for extracting information necessary for rhyme control of a speech synthesizer using the same.

In particular, the present invention targets Korean grammatical relations between sentences in sentences in Korean, for example, sentences with the opposing structure of the housewife (NP: NOUN PHRASE) and the predicate (VP: VERB PHRASE). The purpose of this study is to provide a syntactic analysis means to extract the information necessary for Korean rhyme control from the Korean language, and to approximate the synthesized sound optimally, thereby improving the quality of the synthesized sound.

In addition, an object of the present invention is to provide a speech synthesis system that improves the naturalness and clarity of the synthesized sound while proceeding with the syntax analysis of Korean.

Hereinafter, a syntax analysis apparatus of the present invention and a syntax analysis method performed in the apparatus will be described with reference to FIGS. 2 through 16 of the accompanying drawings.

First, Figure 2 shows the functional configuration of a speech synthesis system including a parser of the present invention.

That is, based on the phonological sentence processing information provided by the sentence preprocessing unit, syntax analysis information (PARSIN information) is extracted, and the segmentation analysis information (according phrase and phrase, due diligence, length of scarcity, etc.) is rhyme implementation and phonological phenomenon processing unit. By supplying to, obtain the form of intonation for the whole sentence and adjust the rest period between words to approximate the synthesized sound to natural sound.

The syntax analysis performed by the parser analyzes the internal structure of the word through preprocessing for each word of the input sentence, and then reveals the grammatical relationship between the previous and the next word.

That is, the parser analyzes the entire syntactic structure as a combination of syntax elements and accordingly identifies the sentence as an opposing structure of the housewife and predicate. He separates the ball, and finds the length of the actual word and the fiction of a word.

The overall configuration of such a parser is shown in FIG. 3, and the overall parse process performed by the parser is shown in FIG.

First, referring to FIG. 3, the syntax analyzing apparatus of the present invention analyzes a structure of a file input means 1 for reading a sentence to be synthesized and a structure of the read sentence through sentence form analysis. Morphological analysis means (2), dictionary means (3) for providing word-by-word analysis information required by the morphological analysis means, and the result of analysis of the morphological analysis means is input to perform syntactic analysis of sentences Syntactic analysis means (4), a syntax rule processing means (5) for performing the processing of the syntax rules required by the syntactic analysis means, and a sentence separation of the sentence by receiving the processing result of the syntactic analysis means. A phrase separation means (6) for performing a; and an accent separation means (7) for separating the accent of the sentence by receiving the processing result of the syntactic analysis means; and Take input And rhyme information extracting means 8 for extracting rhyme information.

The syntax analysis process of the Korean sentence made by the apparatus of the present invention configured as described above uses a small amount of morpheme dictionary to obtain information about a word through morphological analysis of the word and input the information into the morphology. Analysis process and syntax analysis proceeds with the morphology, and the phrase is divided into sentence elements as a result of syntax analysis, and the phrase and intonation are separated and the syntactic analysis process loads the syntax analysis information in sentence units. .

The morphological analysis process includes a first process of inputting a sentence to be synthesized, a second process of dividing a word of the input sentence into a consonant, neutral, and final word, and a dictionary search of the separated word. The third step of constructing the information frame of the word form by revealing the internal structure of each word, and the fourth step of reconstructing the information through error correction by the morphological analysis of the information frame obtained in the above process.

That is, in the morphological analysis process, as shown in FIG. 14, words of the input sentence are cut into initial, neutral, and final words, and a word structure is formed through dictionary search for each word, and errors are corrected through morphological analysis. Restructure the framework.

The syntactic analysis process includes a first process of inputting an information frame of a sentence for a sentence, a second process of performing syntax analysis word processing based on the input information frame, and a sentence from the syntax analysis result. A third process of separating the phrase and the intonation phrase of the second phrase, and adjusting the value of the rhythm processing to be applied according to the number of words and the number of syllables in the separated phrase and the intonation phrase, and outputting the syntax analysis information in the input sentence and outputting it The process takes place.

That is, in the syntactic analysis process, as shown in FIG. 15, the morphological analysis process is performed by receiving the morphology processed in the morphological analysis process, and the phrase and the intonation are separated by the syntax analysis result. PARSING information) is printed in sentence unit.

4 illustrates the overall flow of the parsing process according to the present invention as described above. Referring to this, the parsing process performed in the parsing apparatus of FIG. 3 is as follows.

First, the Korean sentence to be synthesized is read from the file input means 1.

This input file is input to the morphological analysis means 2, and the morphological analysis means cuts the input file by sentence.

In order to analyze the internal structure of the word, information in the dictionary means (3) consisting of the BOUND FORM MORPHEME and the partial FREE FORM MORPHEME of the word or investigation center is read, and each word is read and written In other words, pretreatment is performed by dividing into the finality.

The dictionary search order of the dictionary means 3 is performed in the same order as in FIG. 5 (a), and the dictionary information provided at this time is provided from the dictionary group of the database as shown in FIG. 5 (b).

The sequence of sequence as shown in (a) of FIG. 5 is the order necessary to extract the maximum information from the limited dictionary, and the indicators 1, 2, necessary dictionary, part-of-speech dictionary, mother-dictionary dictionary, part-of-speech dictionary, and mother-search The dictionary search is performed in order of dictionary, part-of-speech dictionary, and shape index dictionary.

In the dictionary means 3, a part-of-speech dictionary group, an element dictionary group, a form index dictionary group, a case and end dictionary group 1, 2, and a section index dictionary group 1, 2 which provide the sentence analysis information are constructed as a database. .

Each dictionary group is a collection of dictionaries made up of the same kind of morphological elements.

The task of extracting information from a dictionary like this is to divide the word into initial, neutral, and final words, match them with the dictionary, and if there is a match, cut out the words as matched in the word and retrieve the information while matching the rest from the next dictionary group. It is divided into two cases.

1) search for independent words

This is a part-word dictionary group or element dictionary group that searches while matching the word values.

2) dependency search

The task is to find information for dividing words into elements by using words or surveys.

For example, if the word 'forward' is cut into initial, neutral, and final, 'ㅇ ㅏ ッ ㅇ ㅡ ㅇ ㄹ ㅇ ㅇ', and in the case of independent word search, the word 'o ㅏ ㄴ ㅇ ㅡ ㅇ ㅇ' In case of dependent search (1). , ㅇ ㅡ ㅇ ㄹ ㅗ ㅇ, (2). O o o o o o l o o (3). o ㅡ o ㄹ ㅗ o, (4). O o d o (5). o r ㅗ o, (6). ㅗ ㅇ ㅇ, (7). ㅗ o, (8). o, in the order of item 1 to item 8 from the center around the back cut one by one to find a match to retrieve information.

In this example, the word 'front' and the search of 'ro' are combined, so the first step is to find 'ro', and the remaining 'o't' is passed to the next dictionary search routine so that 'front' is used as a noun. You will find

Thus, the word of this example completes the search for the dependent word in item 3 above. In this way, when the internal structure of the input sentence is analyzed in the dictionary, the input sentence is input into a form composed of seven cells as shown in the example of FIG.

The information frame of a fish form consists of seven cells (CELL) of an element, a part-of-speech, an index, a grid, a mother, a grid.

The value input to this cell is the eigenvalue of the matched dictionary, and the meaning of these eigenvalues is as shown in (b) of FIG.

In this way, morphological word analysis and construction of the morphology in FIG. 4 are completed. The value of the morphology is then analyzed according to morphological rules to correct wrong cell values and to cut due diligence and vanity.

This completes the morphological analysis of the input sentence by the morphological analysis means 2 of the syntax analysis device of the present invention.

This morphological analysis information is supplied to the syntactic analysis means 4, and the syntactic analysis means performs syntax parsing word processing as shown in FIG. 7 with reference to the syntax rule processing means 5 based on the input information.

The parsing word processing process includes noun ordering processing process (A), predicate processing process (B), concatenated word processing process (C) by connection check, section index processing process (D), and word value. And a step (F) for separating the connection compound statement and a step (G) for separating the compound compound statement. It demonstrates with reference to FIG.

The symbols used in FIGS. 8 to 11 and the meanings of the terms are the same as those in FIG. 16.

1. Noun Listing Process (A)

The noun sequence processing process includes a first step of receiving a word-by-word information frame, a second step of searching whether the input word has a vowel (search, a mother) and an independent morpheme, and the second step of the second step. The third step of searching for a punctuation symbol if the result is a vowel or an independent morpheme, and if the search result punctuation symbol is found in the third step is terminated, returns to the first step and analyzes the next word. A fourth step of searching for a comma if the search result of the second step is not an independent morpheme or a vowel, and if the comma search result does not contain a comma, a phrase is searched for a phrase element If the second word is combined with the current word and returns to the first step, and if the next word is not a syntax element, the sixth step is returned to the first step, and if there is a comma, the previous word is returned. 7th step of searching whether there is a modifier in the case, and if there is a modifier in the result of the modifier search, after returning to the first step after the completion of the routine, the next word is input and the noun is listed if there is no modifier in the previous word. In step 8, the method returns to the first step and receives the next word.

As in Fig. 8, if a word is omitted from a word or is a list of nouns, it is treated as a noun.

For example, if the survey is omitted, it is 'national education site' and if it is a noun, it is 'mountain, sea, river'.

Therefore, as in FIG. 8, when the information frame as shown in FIG. 12 is obtained as a result of construction, the information as shown in FIG.

That is, when the word-specific information frame (TEMPLETE) is input in step (A), the next step (B) checks whether the word (search or ending) is in the word, and also checks whether the input word is an independent morpheme. do.

If the result of the test is not an independent morpheme or a vowel, it checks the comma (step C). If there is a comma, it is determined whether the previous word modifies the current word in step (F). In step H, the routine stops and the next word is analyzed (step 1).

If there is no modifier in the word preceding the search result, the next word is received in step G, and it is determined as a noun and the first routine is repeated.

In step (B), if there is an independent morpheme or toe, check the sentence symbol (SENTENSEMARK) in step (J), and if there is a sentence symbol (step K), if there is no symbol, the next step After (1), the following words are analyzed as above.

On the other hand, if there is no comma in the comma retrieval step (C), it is checked whether or not a syntax element is included in the next word (step D). If the next word is a syntax element, the next word and the current word are bundled and the constant is initialized.

The phrase element search step (D) is to find a toe or a phrase element.

In this way, if the noun processing is completed, the same predicate processing is performed in FIG.

2. Predicate Processing Process (B)

The predicate processing process includes a first step of receiving word-specific information, finding a combination of words and checks of the input word, analyzing a part-of-speech for a word, and classifying verbal forms in sentences as endings and indexes. A second step of determining a third step, and a third step of determining a comma and a punctuation mark after increasing the number of words that are combined and processed as a predicate ending, a connecting ending ending ending, and a dependent ending, if the word is a verbal form. A fourth step of determining a comma and a punctuation symbol after grouping when the result of the determination of the second step is a continuation type or a side word; and a comma and a punctuation symbol when the descriptive judgment result segment indicator of the second step is developed. The fifth step of determining and if the result of the determination of the comma and punctuation symbols is terminated if there is a comma and punctuation symbols, return to the first stage if there is no The sixth step is to analyze words.

This predicate processing is a process of finding a predicate, and a predicate can be composed of several predicates, and in this case, it is treated as a predicate.

For example, the phrase "coming soon" is found in the phrase "coming soon". That is, referring to FIG. 9, the words are input one by one in the information frame (step A), and the judgment of the predicate is performed therefrom.

Judgment of the predicate is to take out the combination of surveys and verbs, to analyze parts of speech for words, and to classify verbal forms in sentences as mothers and indicators.

As a result of judging such a predicate, the word which is a teletype or a local element is transferred to a grouping processing step (M) through a step (CDE), and the word that is a verbal type is a predicate ending, a connecting word, The dependent words are processed (steps F, G, and H), and the bound words increase the number of words (j) and the index (end) indicating whether the processing routine is continued is set to 1 (step I). If so, the routine continues.

Then, the comma or punctuation is searched (step J), and if there is a punctuation, it is terminated. If there is no comma or punctuation, the next word is analyzed.

As a result of the predicate determination, if it is a spoken type or a side word, if it is a routine that continues before (step M), the routine is terminated with end = 0 (step N), and a sentence symbol search step is performed (step J). .

As a result of the above judgment, the division index is searched (step L), the sentence symbol is searched (step J), the next word is analyzed (steps K, B, and C), grouped and grouped (step M), or the routine is skipped. Comes out. In this way, the processing of the predicate is completed and the connection inquiry process is performed.

3. Connection word processing process by connection investigation (C)

The access inquiry processing process includes a first step of receiving an information frame for each word, a second step of searching for an access search from the information frame of the input word, and a sentence symbol for searching for a sentence symbol if there is no connection search result. If there is a search result of the next word, and if there is a search result of the second step, if there is a search result of the second word, it checks whether or not the next word is a subword. A fourth step of checking and a fifth step of increasing the number of words to be bound and continuing the connection check processing routine for the next word if the next word in the fourth step is a word except for a sub-chord; and in the fourth step, If the word is a noun list or if the previous word is a subword and the current word is a subword, the sixth step of returning to the second step after the grouping process is performed to search for the next word; In the fourth step, if it is not a statement, a noun, or a sub-word, a seventh step of changing the connection check to the case check and searching for a sentence symbol.

In this way, the connection lookup process processes the connection word as follows.

Groups of spoken words connected by a connection survey are treated as a single message. Examples of types of connection surveys are wah, gua, lang, me, and ena.

To process the connection of the message by the connection survey, as shown in FIG. 10, the connection survey is first searched for in the input word frame (steps A and B).

If there is no connection check result, the sentence symbol is checked (step D), if there is a sentence symbol (step N), and if there is no sentence symbol, the next word is retrieved through step (S).

If there is a search result as a result of the above search, qbs = 1 enters the connection search processing routine.

At this time, since there is a lot of symmetry in the connection of the Korean word, it is determined whether the subword is a subword in step (E), and if the previous word is a subword, nb = 1 to determine whether the next word is a subword (steps F and G).

If the next word is a statement (step G), nf = 1 (step J), if the nouns are listed (step H), nf = 2 (step K), and if nb = 1, it is a subordinate (step I). , nf = 3 (step L)

In each of the above steps, it is not a statement, a noun sequence, a sub-challenge, or a connection check toned (step M), and it goes to the sentence symbol search step and exits the current routine.

On the other hand, if nf = 1 in each of the above steps (step P), groupings in one routine (step Q) are initialized (step R), and then the next word is input.

If nf is not 1, the number of bound words (nm) is increased (step O), and the connection check processing routine (F, G, H, I, J, K, L, P, Q) for the next word is found. Continue.

In this way, the connecting phrases can be grouped into one word, thereby completing the connection investigation process. When the access investigation process is completed, partial index processing is performed.

4. Sectoral Indicator Process (D)

The division indicator processing process includes a first step of receiving an information frame for each word, a second step of determining an indicator by analyzing the input information frame, and searching for a comma if the determination result is a bore indicator, a sub-index indicator, and an in-use indicator. In the third step and the determination result of the second step, if it is not a bore index, a floating index, or an unused index, the sentence index is searched. If there is a sentence index, the sentence is terminated. In the fourth step of performing the index search and the comma search result in the third step, if the comma is not found, the next word is analyzed and the next word is a floating index, a bore index, and a floating index. After the grouping process is finished, the fifth step is terminated, and if the next word is not a bore mark, a floating mark, or an unmarked mark in the fifth step, the process returns to the second step and returns to the next word. It comprises a sixth step of performing a table search.

The sector indicator processing thus achieved is as follows.

Complex transformations of all sentences can be divided into insertions and links, and the morphemes used in compounding these sentences are sector indicators.

These sector indicators can be used in conjunction with index linkages in compound statements. Their analysis can find insertions and listings of phrases in compound statements.

The type of consolidation of sector indicators is: (1) Bore indicators (eg…). Bore Index Links (about,…), (2) Floating Indicators (to…)… Subsidiary Indicators Linked Words (Restricted,…), (3) Subsidiary Indicators (Sub indicators, Bore Indicators). Bouillon indicator concatenation (because, as, like, ...).

The processing of such sector indicators will be described with reference to FIG.

When a word is input in the input step (A), the search words (B, C, D) are used to find the bore, the float, and the indicator, and enter 1 for each constant (steps E, P, and G).

In the search step, a word that does not correspond to a bore, a floating part, or a bouillon is inputted with the next word (step J, O, P) or completely exits from the routine (step Q).

When the index is searched in the search step, a comma is searched in the comma search step (H) to determine whether the routine continues.

That is, if there is a comma, the routine exits through step J. If there is no comma, i is incremented to analyze the next word in the next processing routine (i ++).

As a result of analyzing the next word, if the next word is a float, bouillon, and bore index and it is connected with the previous processing value (steps K, L, and M), it is grouped in the grouping step (N) and the routine is terminated (step J). ).

In this way, when the division index processing is completed, the word is replaced with the element value (step E), and the section index and the insertion compound statement in the compound statement are cut using the division index (step F, G).

As such, when the syntactic analysis is completed by the syntactic analysis means 4, the phrases are classified into syntax elements and the phrases and intonations are cut by the phrase separation means 6 and the intonation separating means 7.

In other words, the syntax elements are divided into leaf elements and essential elements, and their relations are examined and grouped into accents.

The leaf elements are, for example, tied together with a subchair-subject and a subordinate relationship. 2. Terminology—to be bound by adjectives and adjectives. Essential elements are, for example, 1. subject, 2. object, 3. bore, 4. predicate.

In this way, when the phrase is cut, the rhyme information extraction means 8 changes the rhythm processing value to be applied according to the number of words and syllables in the phrase and intonation. (Change the parsing level) After changing the rhyme processing value, the rhyme information in the sentence is changed. Rhythm control by syntactic analysis of Korean language is completed by outputting as shown in the example of FIG.

In FIG. 12, each value in the information frame is an intrinsic value of a morpheme that constitutes a word. Of these values, negative values are removed after completing the correction of errors by morphological analysis.

According to the grammatical relationship between each word's intrinsic value and surrounding words, the value of the distinguishing element (CV) is obtained, which is the value when the function of the word in the sentence is expressed as the distinguishing element. Each CV value is the same as the prior eigenvalue in (b) of FIG. 6 mentioned above. For example, a CV value of 40 or 49 is both a type of subcarrier.

The result of syntax analysis (RESULT) shows the phrase and accent are separated in the sentence according to the combined form of syntax element value (CV).

The accent is the smallest unit to which the intonation can be applied and is determined according to the grammatical link between words. '2' is the boundary of intonation and '10' is the boundary of passage.

Using the values found above, the parsing result between the phrase, intonation section, and phrase phrase in the rhyme implementation and phonological phenomenon processing unit outputs the synthesis result of the speech synthesis unit as shown in the following example.

"Revival of the # ancestors of the ancestors # today / / # Independence # often # independence / / # out # contribute to humanity # / /"

In the output synthesized speech, interspace space means a boundary between words, # means an accent boundary, and // indicates a boundary between phrases.

By using the processed sentence as described above, it is possible to draw a CONTOUR as shown in FIG. 13 and to adjust the pause between words.

The rules for controlling intonation and word pauses in sentences are as follows:

1. INTONATION; It represents the pitch of the note, which is determined by the pitch of the speech, which depends on the form and syntactic structure of the sentence.

(One). PITCH SENTENSE = PITCH BASE LINE + PITCH INTONATION PATTERN

The accent of the entire sentence is obtained by the above rule, and the PITCH BASE LINE is obtained by the following equation.

(2). PITCH BASE LINE (t) = (EPH-1PH) * t / TLS + 1PH

There is a tendency of RESETTING at the boundary of the passage as shown in (b) of FIG.

(3) According to the intonation, the pitch pattern PITCH PATTENRN is obtained as shown in FIG.

2. PAUSE

(One). Rest between syllables; Decide according to the syllable's initial and finality.

(2) rest periods between words; Depending on the syntactic structure, the values between sentences, words, intonation sections, and words differ.

As a result of extracting the rhyme information and the resulting speech synthesis, the natural and clarity of the synthesized speech is improved compared to the conventional speech synthesis system, and the voice, which is a human acoustic information transmission means implemented by a machine, is closer to that of the human. High quality sound is now possible.

As described above, according to the present invention, since the speech synthesis system performs rhythm control by performing Korean syntax analysis, a high-quality synthesized sound such as a human voice can be obtained, and thus, an ARS (voice answering system) to which the synthesized sound can be applied. B. When applied to an automatic switching system, it is possible to improve the intelligibility of the voice to ensure accurate voice information delivery and to secure high reliability in voice information transmission.

Claims (12)

File input means (1) for reading the sentence to be synthesized, morphological analysis means (2) for analyzing the structure of the read sentence through sentence morphology analysis, and the morphological analysis means Dictionary means (3) for providing word-by-word analysis information; Syntactic analysis means (4) for receiving syntactic analysis of sentences by receiving analysis results of morphological analysis means, and syntax rule processing means (5) for processing syntax rules required by the syntactic analysis means (5). And a phrase separation means (6) which receives the processing result of the syntactic analysis means and performs sentence separation of the sentence, and an intonation that separates the accent of the sentence by receiving the processing result of the syntactic analysis means. And a rhyme information extracting means (8) for extracting rhyme information by receiving means (7) and the phrase separation result and the intonation section separation result. The apparatus of claim 1, wherein the dictionary means (3) comprises a part-of-speech dictionary group, a urea dictionary group, a form index dictionary group, a case, a mother dictionary group, and a sector indicator dictionary group. The morphological analysis process of composing the information frame of the word form through the word-by-word classification and the word-by-word dictionary search of the input sentence, and the phrase obtained by separating the phrase and intonation by inputting the information frame of the process. Syntactic analysis method for rhyme control of speech synthesis system that consists of syntactic analysis process that outputs analysis information by character unit. The method of claim 3, wherein the morphological analysis process comprises: a first process of inputting a sentence to be synthesized; a second process of separating words of the input sentences into initial, neutral, and final words; and the separated words It consists of a third process of constructing the information frame of words by searching the internal structure of each word through a dictionary search of the word, and a fourth process of reconstructing the information frame through error correction by the morphological analysis of the information frame obtained in the above process. Parsing Method for Rhythm Control in Speech Synthesis System. The method of claim 3, wherein the syntactic analysis process comprises: a first process of inputting an information frame of a sentence for a sentence, a second process of performing syntax analysis word processing based on the input information frame; A third step of separating the sentence phrase and intonation phrase from the syntax analysis result, and adjusting the value of the rhythm processing to be applied according to the number of words and the number of syllables in the separated phrase and intonation phrase, and then analyzing the syntax information in the input sentence. A parsing method for rhyme control of a speech synthesis system consisting of a fourth process of outputting. According to claim 3, The dictionary search for each word in the morphological analysis process, Sector index dictionary (independent language), Sector index dictionary (dependent language), Urea dictionary (independent language), Parts of speech dictionary (independent language), ending dictionary Voice synthesis to search in the order of (dependent), part-of-speech dictionary (independent child), ending investigation dictionary (dependent), part-of-speech dictionary (independent), morphological indicator dictionary (dependent), morphological indicator dictionary (dependent) Parsing method for rhyme control of system. The method of claim 3, wherein the information frame of the morphology in the morphological analysis process is composed of seven cells (CELL) of elements, parts of speech, indicators, intervals, endings, sentence marks, and sentence symbols. Parsing Method. 6. The method of claim 5, wherein the second process of syntactic analysis includes a noun sequence processing that treats a word as a noun when a word is omitted or a list of nouns, and a word is consecutively used. Predicate processing to find the predicates that make up the predicate, connection search processing to group the words of the spoken form connected by the connection search and treat them as a single statement, and to process the sector index to find the segment indicators used for the compounding of sentences. Parsing Method for Rhythm Control of Speech Synthesis System. 9. The method of claim 8, wherein the noun sequence processing step comprises: a first step of receiving a word-by-word information frame, and a second step of searching whether the input word has a toe (irradiation, a mother) and an independent morpheme; A third step of searching for a punctuation symbol in the case of the second result of the search result or a separate morpheme, and a first step if not ending in the case of the third search result punctuation symbol. A fourth step of returning and analyzing the next word; a fifth step of searching for a comma if the search result of the second step is not an independent morpheme or a satin; and if the comma search result does not have a comma, If it is a syntax element, if it is a syntax element, it binds the next word and the current word and returns it with the first step. If the next word is not a syntax element, it terminates. If not, it returns to the first step and analyzes the next word. The fourth step, and if the search result of the second step is not an independent morpheme or no vowel, the searcher is a fifth step, and if the comma search result does not have a comma, the search word is a phrase element. If the second word and the current word are combined and returned to the first step, and the second word is not a syntax element, the sixth step is returned to the first step, and if the comma search result includes the comma, In the seventh step of searching, and if the modifier phrase is found as a result of the modifier phrase, after the completion of the routine, the method returns to the first step and receives the next phrase. A parsing method for rhyme control of a speech synthesis system consisting of an eighth step of returning to a step and receiving the next word The method of claim 8, wherein the predicate processing process comprises: a first step of receiving an information frame for each word, finding a combination of the search and the sentence of the input word, analyzing a part-of-speech for the word, The second step of classifying a verbal form in a sentence and judging a descriptive word, and if the word is a verbal word, the word is processed as a descriptive ending, and the comma and sentence symbol are increased after increasing the number of words that are bound and ending. A third step of judging; and a fourth step of judging commas and punctuation symbols after grouping when the result of the judgment of the second step is a spoken type or a side word; Is found, a fifth step of determining a comma and a punctuation symbol; A parsing method for rhyme control of a speech synthesis system, comprising a sixth step of returning to the system and analyzing the next word. 10. The method of claim 8, wherein the access inquiry process comprises: a first step of receiving an information frame for each word and a second step of searching for an access search from an information frame of an input word; The third step of searching for a symbol and ending if there is a punctuation symbol, if not, the third step of searching for a connection check of the next word, and if the search result of the second step has a connection check, checking whether the previous word has a floating type. A fourth step of checking whether the word is a subword, and a fifth step of increasing the number of words to be bundled if the next word is a subword except for the subword in the fourth step, and continuing the connection check processing routine for the next word; In the fourth step, if the next word is a noun or the previous word is a subword and the current word is a subword, the grouping process returns to the second step to check the connection of the next word. Rhythmic parsing method of the speech synthesis system consisting of a sixth step of searching, and a seventh step of changing the connection search to dichotomy and searching for punctuation when the fourth step is not spoken, nouns, or subordinates. . The method of claim 8, wherein the process of processing the sector indicator comprises: a first step of receiving an information frame for each word, a second step of analyzing the input information frame to determine an index, and a bore index and a floating index of the determination result; In the third step of searching for a comma, if the indicator is not a bore indicator, a sub indicator, and a sentence indicator, if it is not the indicator, the sentence indicator is terminated. In the fourth step of performing index search for the next word by returning to the step, and in the third step, if there is no comma as a result of the comma search, the next word is analyzed by using the next word. If the previous word is connected to the above-mentioned processing value, the grouping process is terminated after the grouping process, and if the next word in the fifth step is not a bore mark, a float mark, or a markup mark, it returns to the second stage. And prosody control syntax analysis method of the speech synthesis system of claim 6, wherein performing a search index for the following Eojeol.